Files in this item

FilesDescriptionFormat

application/pdf

application/pdfWANG-DISSERTATION-2016.pdf (3MB)Restricted to U of Illinois
(no description provided)PDF

Description

Title:Leveraging multi-dimensional, multi-source knowledge for user preference modeling and event summarization in social media
Author(s):Wang, Jingjing
Director of Research:Han, Jiawei
Doctoral Committee Chair(s):Han, Jiawei
Doctoral Committee Member(s):Zhai, Chengxiang; Hockenmaier, Julia; Mei, Qiaozhu
Department / Program:Computer Science
Discipline:Computer Science
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:Ph.D.
Genre:Dissertation
Subject(s):user preference
event summarization
Abstract:An unprecedented development of various kinds of social media platforms, such as Twitter, Facebook and Foursquare, has been witnessed in recent years. This huge amount of user generated data are multi-dimensional in nature. Some dimensions are explicitly observed such as user profiles, text of social media posts, time, and location information. Others can be implicit and need to be inferred, reflecting the inherent structures of social media data. Examples include popular topics discussed in Twitter or Facebook, or the geographical clusters based on user check-in activities from Foursquare. It is of great interest to both research communities and commercial organizations to understand such heterogeneous data and leverage available information from multiple dimensions to facilitate social media applications, such as user preference modeling and event summarization. This dissertation first presents a general discriminative learning approach for modeling multi-dimensional knowledge in a supervised setting. A learning protocol is established to model both explicit and implicit knowledge in a unified manner, which applies to general classification/prediction tasks. This approach accommodates heterogeneous data dimensions with a significant boosted expressiveness of existing discriminative learning approaches. It stands out with its capability to model latent features, for which arbitrary generative assumptions are allowed. Besides the multi-dimensional nature, social media data are unstructured, fragmented and noisy. It makes social media data mining even more challenging that a lot of real applications come with no available annotation in an unsupervised setting. This dissertation addresses this issue from a novel angle: external sources such as news media and knowledge bases are exploited to provide supervision. I describe a unified framework which links traditional news data to Twitter and enables effective knowledge discovery such as event detection and summarization.
Issue Date:2016-07-08
Type:Thesis
URI:http://hdl.handle.net/2142/93043
Rights Information:Copyright 2016 Jingjing Wang
Date Available in IDEALS:2016-11-10
Date Deposited:2016-08


This item appears in the following Collection(s)

Item Statistics